2015
Essays on natural resources and labor economics
Mohammad Mainul HoqueIowa State University
Follow this and additional works at:https://lib.dr.iastate.edu/etd
Part of theAgricultural and Resource Economics Commons,Labor Economics Commons, and theNatural Resource Economics Commons
This Dissertation is brought to you for free and open access by the Iowa State University Capstones, Theses and Dissertations at Iowa State University Digital Repository. It has been accepted for inclusion in Graduate Theses and Dissertations by an authorized administrator of Iowa State University Digital Repository. For more information, please contactdigirep@iastate.edu.
Recommended Citation
Hoque, Mohammad Mainul, "Essays on natural resources and labor economics" (2015). Graduate Theses and Dissertations. 14576.
Essays on natural resources and labor economics
by
Mohammad Mainul Hoque
A dissertation submitted to the graduate faculty in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
Major: Economics Program of Study Committee: Catherine L. Kling, Co-Major Professor
Peter F. Orazem, Co-Major Professor Joseph Herriges
John Schroeter Artz Georgeanne
Iowa State University Ames, Iowa
2015
DEDICATION
To my parents, and all taxpayers in the US and Bangladesh who contributed to finance my education from kindergarten to the graduate school.
TABLE OF CONTENTS
ACKNOWLEDGEMENT
………
viGENERAL ABSTRACT
…..………
………..
viii CHAPTER 1. GENERAL INTRODUCTION ………... 1CHAPTER 2. IS OUTDOOR RECREATION RECESSION PROOF? AN EMPIRICAL INVESTIGATION OF THE 2009 RECESSION
………..
41 Introduction………. 4
2 Background and Theoretical Motivation ……….. 8
3 Econometric Framework ……….. 12
3.1 Empirical design and strategy ……….. 12
4 Iowa Lake Survey ……….. 18
5 Results & Discussions ………... 21
5.1 Propensity score estimation……….. 21
5.2 Impact on participation ……… 23
5.3 Impact on total number of trips………. 25
6 Robustness.……….... 26
6.1 Placebo exercise……… 26
6.2 Specification without recreation preference variable ……….. 27
6.3 Matching within RUCA cell ……… 28
7 Extension 29 7.1 Water quality, employment status, and lake recreation behavior during recession ……….. 29
7.2 County unemployment and recreation ………. 31
8 Conclusion………. 34
9 References ………. 35
CHAPTER 3. DOES JOB LOSS DURING RECESSION AFFECT
RECREATIONEXPENDITURE? AN INVESTIGATION INTO PANEL
STUDY OF INCOME DYNAMICS DATA ………... 68
1 Introduction ………... 68
2 Panel Study of Income Dynamics (PSID) Data ……… 71
3 Empirical Design and Strategy ………. 75
3.1 Linear fixed-effect model ……… 76
3.2 Treatment effect framework ……… 78
4 Results and Interpretation ………. 83
4.1 State unemployment and recreation expenditure ………. 83
4.2 Job loss within household during recession and recreation expenditure ………. 85
5 Conclusion ……… 92
6 References ………. 93
APPENDIX B. APPENDIX TO CHAPTER 3 ……… 109
CHAPTER 4. HOW LIFE EXPECTANCY AT BIRTH AFFECTS SCHOOLING INVESTMENT AND LIFETIME EARNINGS: EVIDENCE FROM CROSS-COUNTRY HOUSEHOLD SURVEYS
……….
1191 Introduction ………... 119
2 Theoretical Framework ………. 123
2.1 Comparative dynamics from increase in life expectancy at birth T ……… 127
3 Reduced Form and Econometric Specification ……… 128
4 Data ………... 131
5 Empirical Specification ………. 133
6 Results ………... 135
6.1 Life expectancy at birth and education ……… 135
6.2 Life expectancy at birth and earnings ……….. 136
6.3 Return to schooling ……….. 137
6.4 Robustness ………... 139
7 Extension ……….. 140
7.1 Selection due to participation in the labor force ……….. 143
8 Interpretation and Conclusion ………... 145
9 References ………. 146
APPENDIX C. APPENDIX TO CHAPTER 4 ………... 165
CHAPTER 5. LIFE EXPECTANCY AT BIRTH AND LIFETIME HUMAN CAPITAL INVESTMENT
………..
1801 Introduction ………... 180
3 Theoretical Motivation and Empirical Framework ……….. 185
4 Data ………... 190
5 Results ………... 192
5.1 Survey by survey estimates ……….. 192
5.2 Estimates from the regression on pooled surveys ……… 193
5.3 Heterogeneity across groups ……… 194
5.4 Life expectancy at higher ages ………. 195
5.5 Robustness checks ………... 196
5.6 Extension ………. 199
6 Discussion & Conclusion ………. 200
7 References ………. 202
APPENDIX D. APPENDIX TO CHAPTER 5 ………... 217
CHAPTER 6. ECONOMIC VALUATION OF ECOSYSTEM BENEFITS FROM CONSERVATION PRACTICES TARGETED IN IOWA NUTRIENT REDUCTION STRATEGY 2013: A NON MARKET VALUATION APPROACH
………..
2191 Introduction ……… 219
2 Ecosystem Services from Nutrient Reduction Strategy ……… 220
3 Methodology for this Study ……….. 222
3.1 Quantification of benefits from improved water quality ………. 222
3.2 Quantification of benefits from reduced soil erosion ……….. 224
3.3 Quantification of benefits from wildlife and carbon sequestration ………. 226
4 Benefits from Agricultural Conservation Practices ……….. 227
4.1 Benefits from construction of wetland ………. 227
4.2 Benefits from cover crops ……… 229
4.3 Benefits from land retirement ……….. 229
4.4 Benefits from buffers ………... 230
4.5 Benefits from reduced tillage ……….. 231
4.6 Benefits from nutrient application at MRTN rate ……… 232
4.7 Ecosystem services not quantified ………... 233
5 Water Quality Benefits ………. 235
5.1 Water quality benefits to local homeowners ……… 235
5.2 Recreation benefits ……….. 238
5.3 Water quality and drinking water treatment costs ………... 239
6 Comparison across Conservation Practices in terms of Benefits ………. 245
7 Conclusion ……….... 247
ACKNOWLEDGMENTS
I would like to thank my co-major professors, Catherine Kling and Peter Orazem, for their valuable thoughts and insights and the numerous hours they spent guiding me in the right direction by discussing research, demonstrating the difference between good and bad research, and emphasizing research integrity. I would also like to offer a special thank you to professor Orazem for allowing me to pursue research in development economics, an area that is not of relative strength in the Iowa State Economics Department, but is one of great personal interest. Without professor Orazem’s dedicated involvement in arranging me access to World Bank’s restricted data, I would never have been able to explore such interesting areas in health and economic development. I would like to express my unbounded gratitude to professor Kling for the generous research support she provided during my last year at CARD. Professor Kling introduced me to new potential interdisciplinary research areas concerning the linkage between agriculture and the environment that enriched my exposure and skills. I also appreciate the patience and unconditional assistance with all my limitations professors Kling and Orazem demonstrated throughout my dissertation period. My gratitude extends to professor Joseph Herriges for countless sessions with insightful suggestions that were invaluable to moving this research forward. Thanks to Dr. Artz for advising me on several occasions and providing financial support. I am grateful to my committee members, professors Brent Kreider and John Schroeter, for their valuable comments that helped to improve this research. Fellow graduate students in the department, Donggyu Yi, Yongjie Ji, and Hocheol Jeon were always helpful whenever I wanted to discuss something related to my research. Thanks are due to the Resource and Environmental Policy group in CARD, especially to Dr. Philip Gassman, for providing
useful information and feedback while I was working on chapter six in this dissertation. Thanks to Nathan Cook for editorial services on several occasions.
I would like to thank my lovely wife, Auditya Shamsuddin, who accompanied me to the US, sacrificing her career in Bangladesh, and always stayed by my side during all the pains and struggles I experienced during my PhD journey at Iowa State. I do not have enough words to express my thanks to her unsung contributions. Thanks to my parents, siblings, and parents-in-law for the unconditional love, inspiration, and for believing in me and in my passion for graduate studies. Finally, I want to show my appreciation to my son, Ishaar Hoque, for being a source of constant inspiration since he was born in January, 2014, even though I was depriving you of enough daddy time while I focused on my research and completed this dissertation.
ABSTRACT
This dissertation consists of five empirical chapters spanning the areas of natural resource economics and labor economics. After a general introduction in chapter one, the next four chapters deal with how households respond to exogenous changes to economic opportunities such as shocks to employment or to life expectancy at birth. The fifth chapter investigates the linkage between agricultural management and ecosystem services. The dissertation makes extensive use of household survey data, both from the US and from a large number of cross-country surveys. The first two chapters show that unemployment during recessions may lower households’ recreation expenditure but increase households’ participation in local outdoor recreation activities. The findings from the third and fourth chapters suggest that rising life expectancy at birth increases years in school as well as lifetime earnings, which reinforces the role of health in economic development. The final chapter provides an estimate of the
environmental benefits associated with the set of agricultural conservation practices identified in Iowa nutrient reduction Strategy 2013. The economic value from local recreation and aesthetics, drinking water purification, reduced soil erosion, and reduced greenhouse gas emissions are sizable and under some assumptions are of same order of magnitude as the estimated costs.
CHAPTER 1. GENERAL INTRODUCTION
This dissertation consists of five essays. In the second chapter, we investigate how the recession of 2008-2009 affected Iowans’ outdoor recreation behavior. The U.S. economy was hit hard by a recession during 2008-2009. During periods of high unemployment, many households suffered income losses which resulted in lower spending on normal goods. However, with changes in employment status, members of some households also experienced a lower opportunity cost of time, and may therefore undertake more household activities that are time intensive. The
opposing effects of lower income and cheaper time associated with unemployment motivate the research question of how effects from the recession alter household recreation behavior. To study effects of this type requires detailed household- level data both before and after a recessionary event. In this study, we utilize a panel data set that is uniquely suited to studying the effects of recession on micro decision making in the context of household recreational choices.
Specifically, utilizing a panel comprised of both pre-recession and post-recession data on
household employment status, usage of recreational sites, and a suite of socioeconomic variables, this paper investigates how changes in employment status during the recession alters lake-based recreation demand. The findings suggest that outdoor recreation in Iowa, in general, is recession-proof.
In the third chapter, we investigate the relationship between household recreation expenditure and job loss during a national cyclical downturn.. We utilize the Panel Study of Income Dynamics (PSID), a longitudinal data set with information on household yearly trip and vacation expenditure, that constitutes better socioeconomic and occupation data to model changes in employment status during a recession. We use a household-level fixed effect model
and a difference-in-differences estimator to control for possible selection into unemployment or retirement. Our results suggest that both local economic conditions and individual
unemployment during a recession affect recreation expenditure.
The fourth chapter draws empirical evidence from cross-country household surveys on the relationship between life expectancy at birth, human capital accumulation, and lifetime labor market earnings. Life expectancy at birth has improved dramatically over time and across countries during the last century. In the standard Ben-Porath framework, greater life expectancy should increase human capital investment both by extending the period in which an individual devotes full time to training in school and by increasing the fraction of time devoted to training after starting to work. Both types of training should increase lifetime earnings, as would
extending the number of productive work years. Motivated by Heckman’s (1976) lifecycle analysis, we investigate the causal relationship between life expectancy at birth and years of schooling by exploiting cross-birth-cohort and cross-country variation from a pool of 194
household surveys from 115 countries. We treat the country-cohort life expectancy at birth as the health endowment that parents use to plan out the investments in their children’s' education. A gain of 10 years in life expectancy at birth leads individuals to increase their completed
schooling by 1.1 years and raises lifetime earnings by 1 percent. To put this in perspective, life expectancy at birth in the U.S. rose 28 years from 1880 to 1980, but birth cohorts and years of schooling rose by about 6.5 years. Our estimates suggest that rising life expectancy in the U.S. explains fifty percent of this increase in schooling.
The fifth chapter tests the robustness of the link between life expectancy at birth and time spent in school across 919 cross-country household surveys. In 95 percent of the surveys, the
effect of life expectancy at birth on years of schooling turns out to be positive and statistically significant.
The sixth chapter focuses an economic valuation of ecosystem benefits from nutrient reduction strategies. With the aim of improving water quality, the Iowa Nutrient Reduction Strategy 2013 set a goal of reducing agricultural non-point-source generated nitrogen load by 41 percent and phosphorus load by 29 percent in Iowa’s waterways. Various combinations of nutrient reduction technologies are proposed including widespread adoption of conservation practices in farming, land retirement, and wetland restoration can meet the specified target reduction. This study identifies the dollar value of ecosystem benefits resulting from the conservation practices adopted under each of the scenarios. The ecosystem services generated from the nutrient reduction practices include carbon sequestration, increased opportunity for recreation, reduced cost for drinking water purification, aesthetic value of cleaner lakes and streams, reduced soil erosion, enhanced habitat for wildlife, and increased biodiversity. A conservative monetization of these benefits suggests that the benefits of the nutrient reduction practices can exceed the implementation costs.
CHAPTER 2. IS OUTDOOR RECREATION RECESSION PROOF? AN
EMPIRICAL INVESTIGATION OF THE 2009 RECESSION
1. Introduction
The US economy was hit hard by a long recession during 2008–2009, which is considered the longest and most severe economic crisis since the end of the Great Depression. The recession affected individual economic well-being through unemployment, stock market crashes, and falling real estate prices, all of which generated low consumer confidence. While much is known about the effect of recessions on macro-level variables, much less is known about how the effects of recession alter household-level consumption behavior. Specifically, during periods of high unemployment, many households will experience lower income, which results in lower spending on normal goods. However, with changes in employment status, members of some households will also experience a lower opportunity cost of time, and may therefore undertake more household activities that are time intensive. To study effects of this type requires detailed household-level data both before and after a recessionary event.
In this paper, we utilize a panel data set that is uniquely suited to studying the effects of recession on micro decision making in the context of household recreational choices. Specifically, utilizing a panel from the “Iowa Lakes Project” comprised of both pre-recession and post-recession data on household income, usage of recreational sites, and a suite of socioeconomic variables, this paper investigates how employment status changes during the recession affects lake-based recreation demand.
Quasi-experimental studies for impact evaluation have become popular in the economics literature. However, causality analysis relating a recessionary shock to recreation behavior has not been undertaken. Using the 2009 great recession as a natural and exogenous event, we fill
this gap. To our knowledge this is the first study investigating the effects of a recession on micro decision making in the context of household recreational choices.
Recession can affect an individual’s recreation demand through several opposing paths. Even if employment status remains unchanged during a recession, an individual may demand less recreation due to uncertainty and therefore increase precautionary savings. Both income and the opportunity cost of time can be affected for an individual experiencing an employment change. When recession hits the economy, an individual previously employed full time may get fewer paid work hours or be forced into retirement, resulting in a fall in income. However, this change offers more time for leisure and recreation. A change in employment status during recession, therefore, may influence one’s outdoor recreation demand through two opposing effects: a substitution effect from cheaper time and an income effect from a fall in income. Further, employment change might lead an individual to revise plans for exotic vacations and trips, which, in turn, might increase demand for cheap local recreation activities.
The Outdoor Foundation’s aggregate statistics reveal that, compared to 2008, total
participation in outdoor recreation across the United States increased slightly in the recession year 2009. However, nearly 42% of respondents reported that the recession affected their outdoor recreation participation to some extent. At the state level, Iowans’ lake visitation rates increased in 2009 relative to 2005 (Iowa Lake Survey Report 2011). Almost 60% of Iowans participated in some form of lake-based recreation activities in 2009, taking around fifteen single-day lake trips on average. In contrast, the consumer-expenditure survey statistics show that expenditures on pleasure and non-business traveling declined during the recession year of 2008–2009 (Bureau of Labor Statistics 2012).
The literature investigating the relationship between recession and recreation demand is limited. Utilizing two intercept surveys conducted in 2006 and 2009 on Quandary Peak, a very popular hiking place in southeast Denver, Loomis and Keske (2012) find no significant impact of recession on total number of visits, travel expenditure, and willingness to pay for visits.
However, the respondent groups studied before and after the recession are different. Thus, it is not clear whether the survey respondents experienced any employment or wealth shock during the recessionary period.
The Iowa Lakes Project survey data contains individual recreation demand behavior (participation and number of trips) and employment status both before and during the recession. In this random population survey, a rich set of information on Iowans’ lake visitation patterns at 132 lakes was collected, as were demographics including employment status. The survey has been administered five times in total, including once each in 2005 and 2009. The 2005 and 2009 surveys together provide a panel of 2,773 individuals who are observed both before and during the recession. We exploit this panel to investigate how the individuals who move from full-time employment status in 2005 to part-time employment, unemployment, or retirement status in 2009, change their outdoor lake recreation usage, both at the extensive and intensive margins.
In our setting, the treatment group individuals are those who experience an employment shock during the recession year 2009. Assignment to this treatment group is non-random due to both observable and unobservable factors, also known as a selection problem. We use the propensity score matching (PSM) method (Rosenbaum and Rubin 1983) to address the selection problem. Since we have an individual-level panel, we can control for time-invariant unobservable factors utilizing the methodology of Heckman, Ichimura, and Todd (1997) and Smith and Todd (2005).
Following the non-experimental treatment effect literature, we adopt both semi-parametric cross-sectional and difference-in-difference matching strategies to conduct empirical analysis.
In our empirical design, we define the treatment and control groups based on employment status. Individuals who were employed full time both before and during the recession constitute the controls. Individuals who were employed full time before the recession year but had been unemployed, employed part time, or retired during the recession year form the first treatment group. Since the retired individuals might have different recreation preferences compared to those of the unemployed and part-time employed group, we also consider a treatment group that excludes retired people. In the first stage of the PSM analysis, we estimate an individual’s propensity to experience an employment shock based on pre-recession information on individual demographics and recreation usage. Using the estimated propensity scores, we conduct the treatment effect analysis using both levels, where we compare recreation behavior of treated and matched controls in 2009, as well as differences (i.e., a difference-in-difference approach) to control for time-invariant unobservable factors. We apply five different matching methods to check consistency of the results. As a robustness check, we conduct a placebo exercise, include a subset of covariates to estimate the propensity score, and conduct exact matching based on location.
The main results from this analysis reveal that employment change during a recession does not affect outdoor recreation. Households who became unemployed either did not change or increase participation in outdoor lake recreation during the recession. On the intensive margin, they visited lakes as frequently as before the recession. However, people going into retirement during the 2009 recession did not exhibit any systematic differences in recreation behavior compared to what they would have done were they employed full time. Our placebo exercise confirms that our findings are not driven by a pre-existing differentiated trend for the treatment
and control group. Incorporating county-level unemployment rate as a proxy for aggregate economic condition, we extend the analysis in an individual fixed-effect framework. The results suggest that households residing in counties with high unemployment during a recession
participated more in outdoor lake recreation.
The insensitivity of recreation demand to recession implies that there are stable economic benefits from nature-based economic activities. This finding is of direct policy relevance for decisions by public officials concerning nature-based public amenities. Improving water quality and public facilities appears to provide social benefits that are resilient to recessions - the
stability of returns to this form of public good provision may raise its value relative to other local public goods.
2. Background and Theoretical Motivation
Two important components that determine recreation behavior are income and the opportunity cost of time [Bockstael and Hanemann 1987; Cesario 1976; Englin and Shonkwiler 1995; Feather and Shaw 1999; Larson and Shaikh 2004; McConnell 1992]. Like any other economic good, income determines an individual’s purchasing power of recreation services. If recreation is a normal good, the impact of a rise in income is positive, and vice versa if it is an inferior good. Time spent for recreation services has two components: travel time and time spent on site. Phaneuf and Requate (2013) provides a useful recreation demand model to motivate an individual’s optimization between consumption of non-recreation necessities, and recreation goods and services. The individual is naturally endowed with T units of time, out of which she works for H hours in the market for an hourly wage of 𝑤, and allocates the remaining time, 𝑇 − 𝐻, between recreation(R) and leisure(l) so that her utility from consumption of R, l, and the
determined outside the model independent of choices for R, l, and, z. Formally, the individual wants to maximize the utility function U(z, R, l; q), where q represents taste parameters, subject to two separate constraints
i) Money income constraint: 𝑤𝐻 = 𝑐𝑅 + 𝑧, where c is the $ cost of a trip, and
ii) Time resource constraint: = 𝐻 + 𝑙 + 𝑡 ∗ 𝑅 , where time remaining after work hours,
𝑇 − 𝐻, is used for leisure and recreation, and t is the time cost for consumption of each unit of R.
The individual solves the following 2-constraint, utility maximization problem max
𝑧,𝑅,𝐿,µ,𝜆𝑈(𝑧, 𝑅, 𝑙; 𝑞) + 𝜆(𝑤𝐻 − 𝑐𝑅 − 𝑧) + µ(𝑇 − 𝐻 − 𝑙 − 𝑡𝑅). Manipulation of the first order conditions results in 𝑈𝑅
𝑈𝑍 = 𝑐 +
µ
𝜆𝑡 = 𝑐 + 𝜙𝑡. At the
optimum, the marginal monetary benefit from one unit of recreation trip (𝑈𝑅 𝑈𝑍 =
𝛿𝑧
𝛿𝑅) must equate with the marginal cost (𝑐 + 𝜙𝑡)of the trip. The recreation price consists of an explicit part, c, and an implicit part 𝜙𝑡. Solving the first order conditions with specific functional form for utility would give us demand equation for each of 𝑧∗(𝑐, 𝑡, 𝑤, 𝐻, 𝑇, 𝑞), 𝑅∗(𝑐, 𝑡, 𝑤, 𝐻, 𝑇, 𝑞),
𝑙∗(𝑐, 𝑡, 𝑤, 𝐻, 𝑇, 𝑞) and, 𝜙∗(𝑐, 𝑡, 𝑤, 𝐻, 𝑇, 𝑞).
The model succinctly identifies the possible pathways through which recession might influence recreation demand behavior. If the recession affects the recreationist directly through a reduction in working hours, or job loss, the individual experiences a fall in money income but have more available time for leisure and recreation. Thus, the opportunity cost of time to be spent for recreation (𝜙) decreases. However, an opposing effect takes place through the decrease in working hours and resulting fall in income. These two opposing effects are comparable to
substitution and income effect resulting from a price change. Whether the time effect dominates the income effect will determine the overall effect of recession on recreation demand behavior.
Unemployment or, fall in working hours during recession and resulting income loss might lead an individual to demand more cheap local recreation activities. In modeling recreation demand, the choice set often includes an element “stay-at-home” option [e.g., Egan, Herriges and Kling 2009]. This “stay-at-home” option captures everything outside the model including options for other recreation activities such as exotic vacations or an international trip. If a recreationist has plan for such a trip but experiences a fall in income due to a recession, s/he is less likely to make those expensive tours. In such cases, the “stay-at-home” option becomes less appealing, and might induce an increase in demand for local recreation activities. In the model specified above, this is equivalent to saying that corner solution, i.e., 𝑈𝑅
𝑈𝑍 < 𝑐 + 𝜙𝑡, is less likely to occur.
Similar to the static model above, in a dynamic setting (Hoque, Kling, and Herriges 2013) we showed that it is difficult to predict the change in recreation behavior in response to a recession. During a recession, everyone is subject to uncertainty, and one may experience unemployment and fall in income. In the latter case, individuals try to smooth consumption of leisure and recreation across periods by reallocating time-resource within periods and monetary resources across and within periods. While in the former case, precautionary motive sets in which, depending on risk attitude, may alter one’s demand for recreation in any direction. The combined effect of uncertainty and unemployment may go in any direction, and depends on the relative strength of precautionary motive against the consumption smoothing effects. Individual exposure to recession will vary by type and intensity, and, in accordance, responses to such shocks will vary as well. The implication is that the net effect is ambiguous.
Literature search suggests that Loomis and Keske (2012) is the only study that
investigates the impact of an exogenous shock, such as a recession, on recreation behavior, either empirically or theoretically. Their study relies on two intercept surveys conducted in 2006 and 2009 on a single location-Quandary Peak, a popular hiking spot in Southeast Denver, Colorado. They found no significant changes in average number of visits, visitation expenditure, and willingness to pay across periods. Since they did not observe any significant difference in hikers’ income between the two periods, it is not clear whether the survey respondents in their study experienced any employment shock during the recession. If not, the individuals did not face the tradeoff between time resources and income in choosing recreation demand. A longitudinal recreation data incorporating pre-recession and post-recession period will help to figure out correctly whether an individual was affected by recession, and how the affected individual alters recreation behavior during a recession.
In contrast to the recreation demand literature, studies in other applied microeconomic fields explored changes in economic behavior during a recession. For example, health economics literature demonstrates a negative association between business cycle and mortality [Ruhm 2000; 2005]. Other examples include studies on the relationship between recession and child health care [Dehejia and Lleras-Muney 2004; Baird, Friedman, and Schady 2011], recession and food-at-home [Dave and Kelly 2010]. Many of these studies focus on economic goods that have both monetary prices and time costs, and thus resemble recreation demand to some extent.1
1 Time intensive activities (for example, child care) exhibits interesting implication in the context of recession. A
depressed wage during recession reduces the time cost in taking various caregiving activities such as more preventive health visits, breastfeeding, cooking healthy meals, or improving general cleanliness. However, during such contraction income also falls, which might affect parents’ ability to purchase nutritious food or health augmenting inputs. It seems that two opposing effects work simultaneously: substitution effect from cheaper time, and income effect from fall in income.
3. Econometric Framework
Given the multiple influences of recession on recreation usage in theory, the overall effect is purely an empirical question. We study the impact of employment change during recession on lake recreation in a non-experimental setting. In our case, the treatment group includes those who experience a change in employment status facing a recession, and assignment to this treatment group is non-random. This non-random treatment assignment is also known as a selection problem, which can be due to both observables and unobservable factors. The selection problem can hide the true causal effect of a change in employment status during a recession on recreation behavior, and there might be confounding factors that affect both selection into the treatment (experience of a change in employment status) and the outcome variable (trip taking to lake). Propensity Score Matching (PSM) method, due to Rosenbaum Rubin (1983), is one approach to overcome the selection problem. PSM is widely used in the program evaluation literature
[Dehejia and Wahba 2002; Ravallion 2005; Ravallion and Jalan 2003; List et al. 2003; Ferret and Subervie 2013]. Under certain assumptions, the method solves the problem of missing
counterfactual in non-experimental setting. 2
3.1 Empirical design and strategy
In this study, we investigate empirically how the individual’s trip-taking behavior to lake changes in response to a change in employment status during a recession. The relevant periods for the analysis are 2005, the pre-recession year, and 2009, the recession year. As identified in the theoretical model, there are several ways an individual may respond to a recessionary shock.
2 In the first step of a two-step procedure, the method estimates a propensity score (one’s probability of being
included in the treatment group) for each individual in the treatment and control groups based on observed
covariates, and based on that in the second step it matches the treatment observations with the appropriate control to estimate the impact of treatment.
An individual who used to visit lakes before recession may choose “stay at home” option during the recession depending on how the combination of income effect, substitution effects, and consumption smoothing effect resulting from unemployment, and attitude towards risk and uncertainty works. In contrast, depending on the relative strength of these effects, an individual may switch from “stay at home” option and start visiting lakes during recession.3 There are confounding factors that may affect both of an individual’s chance of experiencing a change in employment status, as well as lake recreation behavior during a recession. For example, facing a depressed wage during the recession, an avid trip-taker might choose voluntary unemployment. To control for such confounding factors, a semi-parametric approach, such as PSM method, seems appealing.
Treatment group:The treatment group consists of the set of individuals whose employment status has changed during the recession. In a recession, a previously fulltime employed
individual might become unemployed, part-time employed, or retired.4 The control group in all cases consists of the recreationists who are full-time employed in both of 2005 and 2009. The three treatment and control groups we study are:
𝑇1𝑖 = {1 𝑖𝑓 "i" is fulltime employed in 2005 but Unemployed/Part time/Retired in 2009 0 𝑖𝑓 "i" is fulltime employed in year 2005 and 2009 . (1)
𝑇2𝑖 = {1 𝑖𝑓 "i" is fulltime employed in 2005 but Unemployed/Part time employed in 2009 0 𝑖𝑓 "i" is fulltime employed in year 2005 and 2009 . (2)
𝑇3𝑖 = { 1 𝑖𝑓 "i" is fulltime employed in 2005 but Retired in 2009
0 𝑖𝑓 "i" is fulltime employed in year 2005 and 2009 . (3)
3 By similar reasoning, a recreationist who do not alter lake recreation at the extensive margin may respond at the
intensive margin by increasing or decreasing number of trips.
4 Recognizing the possible differences between unemployed and retirees, we form three treatment groups including
Outcome variable:Let the treatment group indicator is 𝑇 = {0,1}, and year indicator is t = {𝑌09, 𝑌05}. The first outcome variable of interest is a binary variable, 𝑇𝑟𝑖𝑝𝑇,𝑡, indicating whether an individual in group 𝑇 takes any trip in year 𝑡. The second outcome variable is 𝑁𝑇𝑟𝑖𝑝𝑇,𝑡 which denotes the total number of trips for group 𝑇 in year 𝑡.5
Propensity of experiencing a change in employment status during the recession: We first estimate one’s probability of being unemployed or retired during the 2009 recession. There is no clear set of standards on what variables to include in the propensity score equation. However, program evaluation literature [(Heckman, Ichimura, and Todd 1997; Smith and Todd 2005; and Caliendo and Kopeinig 2008] suggests to incorporate all important and necessary variables that may influence both outcome and treatment variables to reduce bias. Accordingly, economic theory, previous research, and institutional setting can help to characterize the covariates. In our setting, previous research is limited to guide us specify the propensity score equation.
Utilizing the treatment status, as defined in equation (1)-(3), and the covariate 𝑿, we estimate the probability of one’s being unemployed or retired using separate logistic regression models for each treatment group l:
Pr(𝑇𝑙= 1|𝑿) = 𝑃(𝑿) =
exp (𝑿′𝜷)
1+exp (𝑿′𝜷). (4) Identification assumption: once we control for propensity score, 𝑃(𝑿), the treatment and the control groups satisfy the ignorability condition, as stated in equation (5) and (6) below. 𝑇𝑟𝑖𝑝1,𝑌09 , 𝑇𝑟𝑖𝑝0,𝑌09 ┴ 𝑇|𝑃(𝑿). (5)
𝑁𝑇𝑟𝑖𝑝1,𝑌09 , 𝑁𝑇𝑟𝑖𝑝0,𝑌09 ┴ 𝑇|𝑃(𝑿). (6)
5 𝑇𝑟𝑖𝑝
1,𝑌09 indicates whether an individual experiencing a change in employment status in 2009 take any trip at all in the year 2009 while 𝑇𝑟𝑖𝑝0,𝑌09 is the similar indicator for individuals who do not experience any such changes in employment. Similarly, 𝑁𝑇𝑟𝑖𝑝1,𝑌09 denotes total number of trips for treatment group while 𝑁𝑇𝑟𝑖𝑝0,𝑌09 denotes total number of trips for the control group in the recession year 2009.
The above states that conditional on the propensity score, exposure to unemployment or retirement during the recession is independent of contemporaneous recreation outcome. One implication of the ignorability condition is the mean equivalence condition which states that once the propensity score is controlled for, the treatment and the control group have similar distribution for the covariate vector 𝑋: 𝑇┴ 𝑋|𝑃(𝑋). In other words,
𝐸[𝑋|𝑃(𝑋), 𝑇 = 1] = 𝐸[𝑋 |𝑃(𝑋), 𝑇 = 0].6
Since we are interested in estimating the impact of a change in employment status during a recession on outdoor recreation, we need a counterfactual estimate on what the recreationists in the treatment group would have done were they not affected by an employment shock during the recession. The weak ignorability assumptions below, a weaker assumption compared to that stated in equation 5 and 6, imply that conditional on the propensity of being in the treatment group, there is no difference in recreation behavior between the treatment and control absent the treatment occurs. Accordingly, recreation behavior of households in the control group in 2009 will be the counterfactual recreation for households in the treatment group, both at the intensive and extensive margin.
𝐸[𝑇𝑟𝑖𝑝0,𝑌09|𝑃(𝑿), 𝑇 = 1] = 𝐸[𝑇𝑟𝑖𝑝0,𝑌09 |𝑃(𝑿), 𝑇 = 0] = 𝐸[𝑇𝑟𝑖𝑝0,𝑌09|𝑃(𝑿)]. (7) 𝐸[𝑁𝑇𝑟𝑖𝑝0,𝑌09|𝑃(𝑿), 𝑇 = 1] = 𝐸[𝑁𝑇𝑟𝑖𝑝0,𝑌09 |𝑃(𝑿), 𝑇 = 0] = 𝐸[𝑁𝑇𝑟𝑖𝑝0,𝑌09|𝑃(𝑿)]. (8)
Estimation: We estimate the impact of a change in employment status during the recession on
recreation adopting the following average treatment effect on the treated (ATT) estimators- 𝐴𝑇𝑇 ̂𝑒𝑥𝑡𝑒𝑛𝑠𝑖𝑣𝑒 = 1 𝑁𝑇[∑ [𝑇𝑟𝑖𝑝1,𝑌09− ∑ (𝑊 ̂𝑖𝑗) 𝑗∈𝐼0 𝑇𝑟𝑖𝑝̂ ]0,𝑌09 𝑖∈𝐼1∩𝑆𝑝 ] (9) and, 𝐴𝑇𝑇 ̂𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑣𝑒 = 1 𝑁𝑇[∑ [𝑁𝑇𝑟𝑖𝑝1,𝑌09− ∑ (𝑊 ̂𝑖𝑗) 𝑗∈𝐼0 𝑁𝑇𝑟𝑖𝑝̂0,𝑌09 ] 𝑖∈𝐼1∩𝑆𝑝 ]. (10)
6 This is also termed as balancing of covariates, 𝑿, which also indicates the quality of the matching estimator. We
where 𝐼1is the set of treated observations, 𝐼0 is the set of control observations, 𝑆𝑝 is the region of common support, and 𝑁𝑇 is the number of observations who belong to the set 𝐼1∩ 𝑆𝑝. 𝑇𝑟𝑖𝑝̂0,𝑌09 is the matched outcome of control observation for treatment “i”, which actually is constructed as the weighted average of all of the matched non-treatment outcomes. Similarly, 𝑊̂𝑖𝑗 is the weight assigned to each matched control “j” corresponding to the treatment observation “i”. Weight will depend on the distance between the propensity scores of treatment “i” and match “j”, and the number of matches as well. For unmatched observations, weight is zero. We applied five different matching algorithms.7
Difference-in-difference matching to control for time-invariant unobservables: The estimators stated in equation 9 and 10 will give the true estimate of a recessionary change in employment status on one’s recreation behavior if selection into such employment change is due to
observable factors X. However, still there may exist unobservable factors, both time-variant as well as time-invariant in nature, that affect both the likelihood of an individual’s exposure to employment change during a recession as well as her recreation behavior. For example, in our context, geographic factors (such as distance to lake, local amenities, local labor market conditions etc.) might confound the results.
Households residing near lakes but not used to taking any lake trip for recreation before a recession may find it relatively easier and cheaper to take some trips after experiencing an employment change during the recession due to more available time and negligible cost of a local trip. On the other hand, one who is living at a place with no lakes in the surrounding
7 In nearest neighbor matching, for each treatment, we pick the control with the closest propensity score, both with
and without replacement. Nearest five neighbors matching picks the five controls with the closest propensity score. Radius matching: for each exposed individual, we pick all the controls whose propensity score lies within a radius distance of ½ and 1/4th of standard deviation of the estimated propensity score.
amenities, but was used to taking trips before recession may find it relatively expensive to make trips after being affected by employment change during a recession. Let us call the former individual as type A, and the later as type B. Without taking into account the influences of location and distance, if we match a type B treatment with a control that lives in a lake-rich locality and is used to taking lake trips anyway, we will not capture true changes in recreation behavior from change in employment status. Similarly, we may end up matching a type A treatment with controls who are dissimilar in terms of locational attributes. Note that we cannot include all potential time-invariant controls, such as one’s residence amenities or local attributes, in the propensity score estimation stage. Heckman, Ichimura, and Todd (1997), and Smith and Todd (2005) strongly recommends using difference-in-difference approach when geographic and other individual specific fixed factors might play potentially confounding role. Since we have a panel, by applying difference (DID) matching estimators we are able to difference out time-invariant unobservable factors.
Compared to a simple propensity score matching estimator, the DID matching estimator will estimate the treatment effect on the differences of outcome variable, which requires
redefining the outcome variables by taking differences in recreation pattern across pre-recession and recession years.8 Next, the DID estimators for out setting are:
𝐷𝐼𝐷𝑒𝑥𝑡𝑒𝑛𝑠𝑖𝑣𝑒 = 𝐴𝑇𝑇̂𝑌09𝑒𝑥𝑡𝑒𝑛𝑠𝑖𝑣𝑒 − 𝐴𝑇𝑇̂𝑌05𝑒𝑥𝑡𝑒𝑛𝑠𝑖𝑣𝑒 , (11) 𝐷𝐼𝐷𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑣𝑒 = 𝐴𝑇𝑇̂𝑌09𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑣𝑒 − 𝐴𝑇𝑇̂𝑌05𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑣𝑒, (12) where 𝐴𝑇𝑇̂𝑌09
𝑒𝑥𝑡𝑒𝑛𝑠𝑖𝑣𝑒 and 𝐴𝑇𝑇̂𝑌05𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑣𝑒 have similar interpretation as in equation 9 and 10.
8 The difference in participation in lake recreation (extensive margin) for the treatment group is ∆𝑇𝑟𝑖𝑝
1=
𝑇𝑟𝑖𝑝1,𝑌09− 𝑇𝑟𝑖𝑝1,𝑌05 while that for the control group is ∆𝑇𝑟𝑖𝑝0= 𝑇𝑟𝑖𝑝0,𝑌09− 𝑇𝑟𝑖𝑝0,𝑌05. Total number of lake trips (intensive margin) is similarly redefined for the treatment group as ∆N𝑇𝑟𝑖𝑝1= 𝑁𝑇𝑟𝑖𝑝1,𝑌09− 𝑁𝑇𝑟𝑖𝑝1,𝑌05 and for the control group as ∆N𝑇𝑟𝑖𝑝0= 𝑁𝑇𝑟𝑖𝑝0,𝑌09− 𝑁𝑇𝑟𝑖𝑝0,𝑌05.
4. Iowa Lake Survey
In this study we utilize data from the Iowa lake survey, a random population survey, which collects a rich set of information on Iowan’s lake visitation pattern as well as
demographics on gender, age, education, employment status, income, and household composition. The survey has been administered five times in total, once in each of the four consecutive years 2002-2005, and the latest is in 2009.9 The surveys in 2005 and 2009 together comprises a panel of 2773 individuals whom we can observe both before and during the
recession in terms of their recreation behavior (both participation and number of trips) and relevant demographics. We first identify the group of people who have experienced a change in employment status during the recession besides those who have not to construct the treatment and control group for our study. Table one presents the employment information across the year 2005 and 2009.
In the 2005-2009 panel, 64.12% of the respondents provided the employment status information for both years. Approximately 6.5% of the people, who were full time employed in 2005, have reported either unemployment or a fall in working hours in 2009. In addition, 10% of the previously full-time employed people have retired in 2009. In the sample, 32.5% of the respondents (900 individuals) do not provide any information on employment status in 2009, which is quite high compared to similar nonresponse in 2005 (5.27%). Again, 52% of these 900 individuals were full-time employed in the pre-recession year 2009.10 However, a simple mean comparison reveals that total number of trips in 2009 of the group with missing employment
9 The survey in 2009 was sent to 10,000 people out of which 4500 were those who responded to a similar survey
conducted in 2005. The survey response rate in 2009 was around 60%.
10 There is possibility that individuals who have experienced employment shock during the recession are unwilling
to share this information. Since in this study, we construct our sample based on individuals’ employment status, significant numbers of respondents are dropping out due to this missing data on employment.
information is not statistically different from that with non-missing employment information.11 This gives us confidence of not being trapped into sample selection bias due to missing data.
Following the definitions given in equations 1-3, we construct the treatment and control groups for our analysis. Table 2 shows the composition of the control and three treatment groups.12 In the analysis we include those who report at most 52 trips in either of the years.13 Among the three, treatment group 1 is the largest in size consisting of 155 observations in total, as it includes retired people besides unemployed and part-time employed. Information on
participation, average number of trips, and demographics across treatment and control groups are reported in Table 3. Participation on average remains unchanged for the control group people across the years of 2005 and 2009. However, unlike the treatment group 3, treatment group 1 and 2 increase participation in lake recreation in 2009. For total number of trips, the pattern is little different. For the control group, treatment group1, and treatment group 3, mean number of trips fall in 2009 compared to 2005. In contrast, treatment group 2 exhibits an increase in mean number of trips in 2009. This gives us an indication of possible differences in recreation behavior across the retired and unemployed people.
Based on the available information in Iowa Lake Survey, the covariates we include while estimating the propensity score are age, polynomials of age, education, gender, number of children in the household, interaction terms between education, age, and gender, recreation patterns in the previous years, and boat ownership. All covariates assume values from the pre-recession period, the survey round in 2005. Education, age, and their interaction terms are
11 Similar comparison between the two groups in year 2003, 2004, and 2005 reveal the same pattern as well. 12 For comparison, an investigation into a similar panel from 2004-2005 shows that 3.5% of the full time employed
people in 2004 becomes unemployed/ part time employed in the year 2005, and 2.5% of the full-employed people in 2004 have retired in 2005.
13 Restriction of 52 trips in one year is to account for explicit day trips. Because some survey respondents might live
motivated by the earning function estimation in the labor economics literature [Mincer 1974; Heckman, Lochner and Todd 2007].14 We assume that factors that determine one’s earnings are also strong predictors for his/her labor market status as well. An individual with a college degree and considerable experience is less likely to be exposed to an employment shock during the recession compared to an individual of similar experience but with only a high school degree. Iowa lake surveys also contain information on households’ residence county and zip code. We match this with the rural-urban commuting area (RUCA) codes maintained by ERS, USDA to classify households by four different types of residence location.15The bottom panel of Table 3 presents summary statistics on various demographics, residence location, and recreation
preference variables observed in pre-recession year for each of the treatment and control groups. Some potentially useful variables to capture the intensity of exposure to recessionary shock, such as household income and work hours, are excluded due to widespread non-response. The idea of habit persistence [Adamowicz 1994; Moeltner and Englin 2004] suggests that pre-recession recreation behavior might influence one’s recreation choices during the pre-recession. Information on past recreation usage, such as total number of day trips and overnight trips taken in the pre-recession year are chosen to group households with similar preferences for recreation or common interest in activity types. Some lake recreation activities, such as fishing, boating and
14 In the labor economics literature experience is often captured by a quadratic of age variable.
15 Detail documentation o RUCA code are available at
http://www.ers.usda.gov/data-products/rural-urban-commuting-area-codes/documentation.aspx#.Uu8l9fldWlI (last accessed on July20th, 2015). As the site notes: “The
rural-urban commuting area (RUCA) codes, a detailed and flexible scheme for delineating sub-county components of rural and urban areas, have been updated using data from the 2010 decennial census and the 2006-10 American Community Survey (ACS). RUCA codes are based on the same theoretical concepts used by the Office of
Management and Budget (OMB) to define county-level metropolitan and micropolitan areas. We applied similar criteria to measures of population density, urbanization, and daily commuting to identify urban cores and adjacent territory that is economically integrated with those cores.”
hunting are included since these will capture preference similarities among recreationists in more subtle manner.16
5. Results & Discussions
We start with a simple OLS model for each of the three treatment groups.17 The OLS estimates reveal that for treatment group 2(unemployed and part-time employed), treatment status and participation in lake recreation are positively associated, which is statistically significant. In contrast, no such association is observed for treatment group 1 and treatment group 3. However, since we are concerned about selection problem with changes in employment status during the recession, we cannot interpret the estimates from OLS in a causal manner.
5.1 Propensity score estimation
Table 4 reports propensity score estimation results for each of the three treatment groups. For treatment group one, education, number of children, age, interaction between age and
education, rural area residence, participation in fishing and boat activities turn out to be
significant predictor of ones probability of experiencing a change in employment status during a recession. For treatment group 2, number of children, participation in fishing, and total number of trips taken in pre-recession year exhibit statistical significance. Similarly, for treatment group 3, education, interaction between age, education and gender, number of children, rural or small-town residence location, participation in fishing and boat activities turn out to be statistically significant predictors of one’s chance of being retired during the recession.18
16 Boating is a dummy variable which assumes value of 1 if a household owns a boat or participates in any of these
boating activities such as jet skiing, canoeing, boating and sailing.
17 In OLS exercises, the outcome variables are participation and total number of trips taken in 2009 whereas the
explanatory variables include the same set of variables used in the matching exercises in addition to the treatment group indicator. We do not report the OLS results here to save space.
18 Since the purpose of these regression estimates is to obtain propensity scores, based on which we will conduct
Based on the estimated propensity score, in each of the cases, we match the treatment with the control applying five matching algorithms (i) nearest neighbor matching without replacement, (ii) nearest neighbor with replacement, (iii) nearest 5 neighbors, (iv) radius matching within a caliper of 1/4th of standard deviation of propensity score, and (v) radius matching within a caliper of 1/2 of standard deviation of propensity score.19 For each matching algorithm, the balancing of covariates is assessed based on two criterions: (i) the difference between mean of treated and matched control group, and (ii) standardized mean difference of the covariates between the treatment and control group.20 Prior to matching, as seen in Table A1, statistically significant differences across the treatment and control group are common. After matching is completed, the covariates balance well. As reported in Table A2-A4 in the appendix A, across treatment groups and matching algorithms, after a matching is conducted, more than 99.5% covariate balance well.21
To satisfy the overlapping condition, while estimating the treatment effects we exclude the treatments that are out of the common support. Table E1 in the appendix A shows the number of matched as well as non-matched treatment in each matching process for each of the three treatment groups. In 95% or more cases, treatments lie in common support region, or, find a comparable counterfactual from the control group.22
19 All of the matching estimation is conducted utilizing package “psmatch2” in STATA 12. 20 The standardized difference of means is calculated as: = 𝑀𝑒𝑎𝑛𝑇𝑟𝑒𝑎𝑡𝑒𝑑−𝑀𝑒𝑎𝑛𝑐𝑜𝑛𝑡𝑟𝑜𝑙
√1
2∗(𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒𝑡𝑟𝑒𝑎𝑡𝑒𝑑 +𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒𝑐𝑜𝑛𝑡𝑟𝑜𝑙)
. Following Rosenbaum and Rubin (1985) we consider a standardized difference of means of 20 as large.
21 For the treatment group two, one covariate, gender, did not balance when nearest neighbor matching is conducted.
Similarly, for the retired group, the covariate education exhibit large standardized difference in the case of nearest neighbor matching without replacement.
22 For treatment group 2 and 3, total 3 treated observations, while for treatment group 1, 2 to 5 treatment
5.2Impact on participation
Table 5 presents the impact of change in employment status during a recession on participation in lake recreation. Note that a simple mean comparison across the treatment and unmatched control shows statistically no significant difference in lake recreation at the extensive margin. For treatment group 1, out of the five matching estimators used, two (nearest five
neighbors matching and radius matching within 0.5 SD of propensity score) show that the treatment group participates more in outdoor lake recreation during the recession. These estimates suggest that households who become unemployed or retired during the recession are 8.3-10.9 percentage points more likely to participate in at least one lake-trip compared to the households who remain full-time employed across the pre-recession and recession period.23
The retired individuals may have distinct recreation preference compared to the
unemployed. Panel b in Table 5 reveals that for the treatment group with employed and part-time employed people, five matching techniques indicate statistically significant positive impact of unemployment during the recession on participation in lake recreation. The estimates imply that an average household that was employed in 2005 but become unemployed in 2009 is 14 to 25 percentage points more likely to recreate in any of the Iowa lakes compared to what s/he would have done if were still full-time employed during the recession year. The bottom panel in Table 5, panel c, reports the results for the retired people. All of the five ATT estimates turn out to be statistically insignificant, which suggests that people who become retired during the recession do not start participating more in lake recreation. Note that in the analysis with treatment group two,
23 In this paper we report the treatment effect (ATT) is statistical significant only if the p value is at least less than or
equal to 0.1. In calculating the p-value, the standard errors are constructed based on 1000 replication of
bootstrapping sample. Each bootstrap sample calculates the propensity score and matching in that sample is done based on that score.
all of the mean differences between the treatment and the control group are bigger in size compared to those we observe for treatment group one. A comparison of estimates across the three treatment groups suggests that the statistically significant impact we obtain for treatment group 1 is driven by the stronger and larger effect from the unemployed and part-time employed group, i.e., treatment group 2.
In the matching results discussed above, although we assume no selection on observables, there can still be unobservable time-invariant confounding factors hiding the true causal
relationship between employment change during a recession and participation in lake recreation, for example distance of lakes from one’s residence. In difference-in-difference matching, we will use the information on a household’s participation in lake recreation both before and during the recession, which will net-out the effects of such time-invariant unobservable factors.
The difference-in-difference matching results for participation are presented in last two columns in Table 5. For treatment group 1, the results are similar to those for participation on the level. From nearest five neighbors and radius matching difference-in-difference estimates, we notice that households who experience a change in employment status during the recession take more lake visits. When we exclude the retired group and conduct difference-in-difference matching on unemployed and part-time employed people only (treatment group 2), all nearest neighbor matching processes show significant positive impact. However, in contrast to the case of matching on the level, radius matching algorithms do not show statistical significance. The bottom panel in Table 5 depicts that none of the matching processes indicate statistically any significant impact of retirement during recession on participation in lake recreation.
This positive effect on participation in lake recreation during the recession by households in treatment group 2 might be attributed to a couple of factors discussed in section 2 and 3. These
households may consider lake recreation as an inferior good, or may have switched from stay at home option to outdoor lake recreation due to reduced opportunity cost of time. But we cannot exactly disentangle which factors are working and to what extent.
5.3 Impact on total number of trips
From our arguments presented in sections on literature review and theoretical motivations, we infer that total number of trips may increase, decrease, or remain unchanged. However, in the propensity score matching analysis, none of the treatment groups show any significant impact of employment change during the recession on total number of trips. Table 6 reports the findings. Although the differences across the treatment and control group are not statistically significant, the positive estimates of average treatment effect on the treated indicates that mean number of trips for the treatment group one and two are higher compared to their counterfactual number of trips. In contrast, the negative estimates for retired households indicate that their total number of lake visits during a recession is lower compared to their counterfactual frequency of visits.
Similar to the arguments presented for participation, we suspect the confounding effects from unobservable factors for total trips as well. To wipe out the mean effects from individually varying but time-invariant unobservable factors, we conduct DID matching estimator. The last two columns in Table 6 report the DID matching results. For any of the three treatment groups, DID estimators does not show any statistically significant impact of employment change during the 2009 recession on frequencies of outdoor lake trips. The DID matching does not change this pattern that we observe for matching on the level.
Contrary to the case for participation, unemployed or partially unemployed households do not increase frequencies of lake trips during the recession. Similar to the analysis for participation, the estimates do not suggest any impact of retirement during the recession on the total number of
trips. However, although the estimates are statistically insignificant, retired households take outdoor lake trips at low frequencies compared to its counterfactual outcome. The finding is consistent across the matching estimators used.
6. Robustness
We conduct three robustness checks. First, we use a placebo recession year to check if our general assumption of no differential trend for treatment and control groups for DID matching estimator is valid in our setting. Second, we change the specification for propensity score estimation including a subset of covariates previously used: we exclude recreation preference variables. Third, we match each treatment observation with controls from the same geographic region to control for unobservable factors that are time-variant in a spatial manner.24
6.1 Placebo exercise
The objective of the placebo exercise is to check whether it is unemployment during the recession or some pre-existing unobservable factors working differently across the treatment and control group are driving our results. If the treatment and control group exhibit differentiated trend in the pre-recession years, and recession truly has no impact on recreation, the DID matching estimator picks up this difference in trend as impact of the change in employment during recession. For the placebo exercise, we assume year 2003 as placebo recession year.25 Table B1-B3 in the
appendix A report that balancing of covariate is satisfied in all cases except for only one covariate
in one matching process for treatment group one. We report the estimates from the Placebo exercise in Table 7 and 8. In all matching processes, neither participation nor frequencies of trips
24 Rural and urban areas may be affected differently during a recession.
25Although Iowa lake project survey was conducted in 2003 and 2004 as well, we have a matched non-missing sample
for all of our treatment and control group observations in year 2003. In the survey year 2004 , we have missing information for 8 treatments and 52 controls from the sample of 971 observations that we are using for our base estimation
in lake recreation turned out to be statistically different across the treatment and control groups in 2003. This finding gives us confidence in saying that our analysis based on matching exercises as reported in the previous section are not contaminated due to differential group trend.
6.2 Specification without recreation preference variable
We estimate the propensity score excluding the recreation preference variables and including only demographics and type of region for residence in the pre-recession year. Table 9 reports the estimates for participation.26 For treatment group 1 and 3, the treatment effect estimates on participation in lake recreation follows the same pattern that we observe previously in Table 5. For these two treatment groups, the difference-in-difference matching estimates are also robust to this different set of observables covariates. Panel b in Table 9 reveals that unemployed and part-time employed households (treatment group 2) participate more in lake recreation during the recession compared to what they would have done had they been employed. Note that, for the DID matching, previously in Table 5, radius matching estimators did not show any statistical significance but under the new setting, all five matching algorithms exhibit statistical significance. For matching on the level of participation, only radius matching estimators show statistical significance whereas previously in Table 5 all five matching estimators turned out to be statistically significant.
For the frequency of lake trips, the results are reported in Table 10. With the new set of covariates, none of the matching estimators across the three treatment groups exhibit statistically any significant effect of a change in employment status during recession on frequencies of trips. Our previous finding that frequencies of lake trips do not change due to unemployment or retirement during the recession is robust to the choice of covariates.
26 Covariates balancing results, as reported in table C1-C3 in the appendix, reveal that quality of the match is good.
6.3 Matching within RUCA cell
Although we have accounted for the effects of mean time-invariant unobservable through matching on the differences, we recognize that we still might end up finding estimates confounded by unobservable fixed factors that vary across regions with time. For example, rural and metropolitan areas may be affected differently during a recession year and exhibit different economic environment. Employment statistics in a rural agricultural county may not change during the recession while employment in the metropolitan area usually drops sharply during the economic crisis. Although we incorporate information on one’s residence location while estimating the propensity score, we still may end up matching a rural treatment with an urban control. Our DID matching estimators cannot control for such region-specific time-variant unobservable confounding factors. So matching individuals within region can control for such time-variant confounding effects.
To control for such possible regionally time-variant confounders, we match each treatment observation with controls from the same RUCA region. An individual from a small town experiencing employment shock during the recession is matched with counterfactuals from a small town area rather than from a metropolitan or rural area. Since we will match exactly within RUCA cell, in the first step, we estimate propensity score excluding variables on geographic regions. The results are reported in Table 11 and Table 12. Table D1-D3 report covariate balance for the cell matching. In contrast to the previous exercises, quality of the matches is not satisfactory here for the treatment group one and three since some covariates do not balance after matching. However, covariates balance well for the treatment group two- the unemployed group.
The estimates in Table11 reveals that when matching is done within the RUCA cell, only one out of five matching estimators for treatment group 1, and none for treatment group 3 show
